Skip to content

models

Classes

fastvideo.configs.models.DiTConfig dataclass 

DiTConfig(arch_config: DiTArchConfig = DiTArchConfig(), prefix: str = '', quant_config: QuantizationConfig | None = None)

Bases: ModelConfig

Functions

fastvideo.configs.models.DiTConfig.add_cli_args staticmethod 
add_cli_args(parser: Any, prefix: str = 'dit-config') -> Any

Add CLI arguments for DiTConfig fields

Source code in fastvideo/configs/models/dits/base.py 
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
@staticmethod
def add_cli_args(parser: Any, prefix: str = "dit-config") -> Any:
    """Add CLI arguments for DiTConfig fields"""
    parser.add_argument(
        f"--{prefix}.prefix",
        type=str,
        dest=f"{prefix.replace('-', '_')}.prefix",
        default=DiTConfig.prefix,
        help="Prefix for the DiT model",
    )

    parser.add_argument(
        f"--{prefix}.quant-config",
        type=str,
        dest=f"{prefix.replace('-', '_')}.quant_config",
        default=None,
        help="Quantization configuration for the DiT model",
    )

    return parser

fastvideo.configs.models.VAEConfig dataclass 

VAEConfig(arch_config: VAEArchConfig = VAEArchConfig(), load_encoder: bool = True, load_decoder: bool = True, tile_sample_min_height: int = 256, tile_sample_min_width: int = 256, tile_sample_min_num_frames: int = 16, tile_sample_stride_height: int = 192, tile_sample_stride_width: int = 192, tile_sample_stride_num_frames: int = 12, blend_num_frames: int = 0, use_tiling: bool = True, use_temporal_tiling: bool = True, use_parallel_tiling: bool = True, use_temporal_scaling_frames: bool = True)

Bases: ModelConfig

Functions

fastvideo.configs.models.VAEConfig.add_cli_args staticmethod 
add_cli_args(parser: Any, prefix: str = 'vae-config') -> Any

Add CLI arguments for VAEConfig fields

Source code in fastvideo/configs/models/vaes/base.py 
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
@staticmethod
def add_cli_args(parser: Any, prefix: str = "vae-config") -> Any:
    """Add CLI arguments for VAEConfig fields"""
    parser.add_argument(
        f"--{prefix}.load-encoder",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.load_encoder",
        default=VAEConfig.load_encoder,
        help="Whether to load the VAE encoder",
    )
    parser.add_argument(
        f"--{prefix}.load-decoder",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.load_decoder",
        default=VAEConfig.load_decoder,
        help="Whether to load the VAE decoder",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-min-height",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_min_height",
        default=VAEConfig.tile_sample_min_height,
        help="Minimum height for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-min-width",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_min_width",
        default=VAEConfig.tile_sample_min_width,
        help="Minimum width for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-min-num-frames",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_min_num_frames",
        default=VAEConfig.tile_sample_min_num_frames,
        help="Minimum number of frames for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-stride-height",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_stride_height",
        default=VAEConfig.tile_sample_stride_height,
        help="Stride height for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-stride-width",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_stride_width",
        default=VAEConfig.tile_sample_stride_width,
        help="Stride width for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-stride-num-frames",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_stride_num_frames",
        default=VAEConfig.tile_sample_stride_num_frames,
        help="Stride number of frames for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.blend-num-frames",
        type=int,
        dest=f"{prefix.replace('-', '_')}.blend_num_frames",
        default=VAEConfig.blend_num_frames,
        help="Number of frames to blend for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.use-tiling",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.use_tiling",
        default=VAEConfig.use_tiling,
        help="Whether to use tiling for VAE",
    )
    parser.add_argument(
        f"--{prefix}.use-temporal-tiling",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.use_temporal_tiling",
        default=VAEConfig.use_temporal_tiling,
        help="Whether to use temporal tiling for VAE",
    )
    parser.add_argument(
        f"--{prefix}.use-parallel-tiling",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.use_parallel_tiling",
        default=VAEConfig.use_parallel_tiling,
        help="Whether to use parallel tiling for VAE",
    )

    return parser

Modules

fastvideo.configs.models.dits

Classes

fastvideo.configs.models.dits.Cosmos25VideoConfig dataclass 
Cosmos25VideoConfig(arch_config: DiTArchConfig = Cosmos25ArchConfig(), prefix: str = 'Cosmos25', quant_config: QuantizationConfig | None = None)

Bases: DiTConfig

Configuration for Cosmos 2.5 video generation model.

fastvideo.configs.models.dits.LongCatVideoConfig dataclass 
LongCatVideoConfig(arch_config: DiTArchConfig = LongCatVideoArchConfig(), prefix: str = 'longcat', quant_config: QuantizationConfig | None = None)

Bases: DiTConfig

Main configuration for LongCat Video DiT.

Modules

fastvideo.configs.models.dits.base
Classes
fastvideo.configs.models.dits.base.DiTConfig dataclass 
DiTConfig(arch_config: DiTArchConfig = DiTArchConfig(), prefix: str = '', quant_config: QuantizationConfig | None = None)

Bases: ModelConfig

Functions
fastvideo.configs.models.dits.base.DiTConfig.add_cli_args staticmethod 
add_cli_args(parser: Any, prefix: str = 'dit-config') -> Any

Add CLI arguments for DiTConfig fields

Source code in fastvideo/configs/models/dits/base.py 
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
@staticmethod
def add_cli_args(parser: Any, prefix: str = "dit-config") -> Any:
    """Add CLI arguments for DiTConfig fields"""
    parser.add_argument(
        f"--{prefix}.prefix",
        type=str,
        dest=f"{prefix.replace('-', '_')}.prefix",
        default=DiTConfig.prefix,
        help="Prefix for the DiT model",
    )

    parser.add_argument(
        f"--{prefix}.quant-config",
        type=str,
        dest=f"{prefix.replace('-', '_')}.quant_config",
        default=None,
        help="Quantization configuration for the DiT model",
    )

    return parser
fastvideo.configs.models.dits.cosmos2_5
Classes
fastvideo.configs.models.dits.cosmos2_5.Cosmos25ArchConfig dataclass 
Cosmos25ArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), _fsdp_shard_conditions: list = (lambda: [is_transformer_blocks])(), _compile_conditions: list = list(), param_names_mapping: dict = (lambda: {'^net\\.x_embedder\\.proj\\.1\\.(.*)$': 'patch_embed.proj.\\1', '^net\\.t_embedder\\.1\\.linear_1\\.(.*)$': 'time_embed.t_embedder.linear_1.\\1', '^net\\.t_embedder\\.1\\.linear_2\\.(.*)$': 'time_embed.t_embedder.linear_2.\\1', '^net\\.t_embedding_norm\\.(.*)$': 'time_embed.norm.\\1', '^net\\.crossattn_proj\\.0\\.weight$': 'crossattn_proj.0.weight', '^net\\.crossattn_proj\\.0\\.bias$': 'crossattn_proj.0.bias', '^net\\.blocks\\.(\\d+)\\.self_attn\\.q_proj\\.(.*)$': 'transformer_blocks.\\1.attn1.to_q.\\2', '^net\\.blocks\\.(\\d+)\\.self_attn\\.k_proj\\.(.*)$': 'transformer_blocks.\\1.attn1.to_k.\\2', '^net\\.blocks\\.(\\d+)\\.self_attn\\.v_proj\\.(.*)$': 'transformer_blocks.\\1.attn1.to_v.\\2', '^net\\.blocks\\.(\\d+)\\.self_attn\\.output_proj\\.(.*)$': 'transformer_blocks.\\1.attn1.to_out.\\2', '^net\\.blocks\\.(\\d+)\\.self_attn\\.q_norm\\.weight$': 'transformer_blocks.\\1.attn1.norm_q.weight', '^net\\.blocks\\.(\\d+)\\.self_attn\\.k_norm\\.weight$': 'transformer_blocks.\\1.attn1.norm_k.weight', '^net\\.blocks\\.(\\d+)\\.self_attn\\.q_norm\\._extra_state$': 'transformer_blocks.\\1.attn1.norm_q._extra_state', '^net\\.blocks\\.(\\d+)\\.self_attn\\.k_norm\\._extra_state$': 'transformer_blocks.\\1.attn1.norm_k._extra_state', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.q_proj\\.(.*)$': 'transformer_blocks.\\1.attn2.to_q.\\2', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.k_proj\\.(.*)$': 'transformer_blocks.\\1.attn2.to_k.\\2', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.v_proj\\.(.*)$': 'transformer_blocks.\\1.attn2.to_v.\\2', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.output_proj\\.(.*)$': 'transformer_blocks.\\1.attn2.to_out.\\2', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.q_norm\\.weight$': 'transformer_blocks.\\1.attn2.norm_q.weight', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.k_norm\\.weight$': 'transformer_blocks.\\1.attn2.norm_k.weight', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.q_norm\\._extra_state$': 'transformer_blocks.\\1.attn2.norm_q._extra_state', '^net\\.blocks\\.(\\d+)\\.cross_attn\\.k_norm\\._extra_state$': 'transformer_blocks.\\1.attn2.norm_k._extra_state', '^net\\.blocks\\.(\\d+)\\.mlp\\.layer1\\.(.*)$': 'transformer_blocks.\\1.mlp.fc_in.\\2', '^net\\.blocks\\.(\\d+)\\.mlp\\.layer2\\.(.*)$': 'transformer_blocks.\\1.mlp.fc_out.\\2', '^net\\.blocks\\.(\\d+)\\.adaln_modulation_self_attn\\.1\\.(.*)$': 'transformer_blocks.\\1.adaln_modulation_self_attn.1.\\2', '^net\\.blocks\\.(\\d+)\\.adaln_modulation_self_attn\\.2\\.(.*)$': 'transformer_blocks.\\1.adaln_modulation_self_attn.2.\\2', '^net\\.blocks\\.(\\d+)\\.adaln_modulation_cross_attn\\.1\\.(.*)$': 'transformer_blocks.\\1.adaln_modulation_cross_attn.1.\\2', '^net\\.blocks\\.(\\d+)\\.adaln_modulation_cross_attn\\.2\\.(.*)$': 'transformer_blocks.\\1.adaln_modulation_cross_attn.2.\\2', '^net\\.blocks\\.(\\d+)\\.adaln_modulation_mlp\\.1\\.(.*)$': 'transformer_blocks.\\1.adaln_modulation_mlp.1.\\2', '^net\\.blocks\\.(\\d+)\\.adaln_modulation_mlp\\.2\\.(.*)$': 'transformer_blocks.\\1.adaln_modulation_mlp.2.\\2', '^net\\.blocks\\.(\\d+)\\.layer_norm_self_attn\\._extra_state$': 'transformer_blocks.\\1.norm1.norm._extra_state', '^net\\.blocks\\.(\\d+)\\.layer_norm_cross_attn\\._extra_state$': 'transformer_blocks.\\1.norm2.norm._extra_state', '^net\\.blocks\\.(\\d+)\\.layer_norm_mlp\\._extra_state$': 'transformer_blocks.\\1.norm3.norm._extra_state', '^net\\.final_layer\\.linear\\.(.*)$': 'final_layer.proj_out.\\1', '^net\\.final_layer\\.adaln_modulation\\.1\\.(.*)$': 'final_layer.linear_1.\\1', '^net\\.final_layer\\.adaln_modulation\\.2\\.(.*)$': 'final_layer.linear_2.\\1'})(), reverse_param_names_mapping: dict = dict(), lora_param_names_mapping: dict = (lambda: {'^transformer_blocks\\.(\\d+)\\.attn1\\.to_q\\.(.*)$': 'transformer_blocks.\\1.attn1.to_q.\\2', '^transformer_blocks\\.(\\d+)\\.attn1\\.to_k\\.(.*)$': 'transformer_blocks.\\1.attn1.to_k.\\2', '^transformer_blocks\\.(\\d+)\\.attn1\\.to_v\\.(.*)$': 'transformer_blocks.\\1.attn1.to_v.\\2', '^transformer_blocks\\.(\\d+)\\.attn1\\.to_out\\.(.*)$': 'transformer_blocks.\\1.attn1.to_out.\\2', '^transformer_blocks\\.(\\d+)\\.attn2\\.to_q\\.(.*)$': 'transformer_blocks.\\1.attn2.to_q.\\2', '^transformer_blocks\\.(\\d+)\\.attn2\\.to_k\\.(.*)$': 'transformer_blocks.\\1.attn2.to_k.\\2', '^transformer_blocks\\.(\\d+)\\.attn2\\.to_v\\.(.*)$': 'transformer_blocks.\\1.attn2.to_v.\\2', '^transformer_blocks\\.(\\d+)\\.attn2\\.to_out\\.(.*)$': 'transformer_blocks.\\1.attn2.to_out.\\2', '^transformer_blocks\\.(\\d+)\\.mlp\\.(.*)$': 'transformer_blocks.\\1.mlp.\\2'})(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (SLIDING_TILE_ATTN, SAGE_ATTN, FLASH_ATTN, TORCH_SDPA, VIDEO_SPARSE_ATTN, VMOBA_ATTN, SAGE_ATTN_THREE, SLA_ATTN, SAGE_SLA_ATTN), hidden_size: int = 0, num_attention_heads: int = 16, num_channels_latents: int = 0, in_channels: int = 16, out_channels: int = 16, exclude_lora_layers: list[str] = (lambda: ['embedder'])(), boundary_ratio: float | None = None, attention_head_dim: int = 128, num_layers: int = 28, mlp_ratio: float = 4.0, text_embed_dim: int = 1024, adaln_lora_dim: int = 256, use_adaln_lora: bool = True, max_size: tuple[int, int, int] = (128, 240, 240), patch_size: tuple[int, int, int] = (1, 2, 2), rope_scale: tuple[float, float, float] = (1.0, 3.0, 3.0), concat_padding_mask: bool = True, extra_pos_embed_type: str | None = None, use_crossattn_projection: bool = False, crossattn_proj_in_channels: int = 100352, rope_enable_fps_modulation: bool = True, qk_norm: str = 'rms_norm', eps: float = 1e-06)

Bases: DiTArchConfig

Configuration for Cosmos 2.5 architecture (MiniTrainDIT).

fastvideo.configs.models.dits.cosmos2_5.Cosmos25VideoConfig dataclass 
Cosmos25VideoConfig(arch_config: DiTArchConfig = Cosmos25ArchConfig(), prefix: str = 'Cosmos25', quant_config: QuantizationConfig | None = None)

Bases: DiTConfig

Configuration for Cosmos 2.5 video generation model.

fastvideo.configs.models.dits.longcat

LongCat Video DiT configuration for native FastVideo implementation.

Classes
fastvideo.configs.models.dits.longcat.LongCatVideoArchConfig dataclass 
LongCatVideoArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), _fsdp_shard_conditions: list = (lambda: [is_longcat_blocks])(), _compile_conditions: list = (lambda: [is_longcat_blocks])(), param_names_mapping: dict = (lambda: {'^x_embedder\\.(.*)$': 'patch_embed.\\1', '^t_embedder\\.mlp\\.0\\.(.*)$': 'time_embedder.linear_1.\\1', '^t_embedder\\.mlp\\.2\\.(.*)$': 'time_embedder.linear_2.\\1', '^y_embedder\\.y_proj\\.0\\.(.*)$': 'caption_embedder.linear_1.\\1', '^y_embedder\\.y_proj\\.2\\.(.*)$': 'caption_embedder.linear_2.\\1', '^blocks\\.(\\d+)\\.adaLN_modulation\\.1\\.(.*)$': 'blocks.\\1.adaln_linear_1.\\2', '^blocks\\.(\\d+)\\.mod_norm_attn\\.(.*)$': 'blocks.\\1.norm_attn.\\2', '^blocks\\.(\\d+)\\.mod_norm_ffn\\.(.*)$': 'blocks.\\1.norm_ffn.\\2', '^blocks\\.(\\d+)\\.pre_crs_attn_norm\\.(.*)$': 'blocks.\\1.norm_cross.\\2', '^blocks\\.(\\d+)\\.attn\\.qkv\\.(.*)$': 'blocks.\\1.self_attn.qkv_fused.\\2', '^blocks\\.(\\d+)\\.attn\\.proj\\.(.*)$': 'blocks.\\1.self_attn.to_out.\\2', '^blocks\\.(\\d+)\\.attn\\.q_norm\\.(.*)$': 'blocks.\\1.self_attn.q_norm.\\2', '^blocks\\.(\\d+)\\.attn\\.k_norm\\.(.*)$': 'blocks.\\1.self_attn.k_norm.\\2', '^blocks\\.(\\d+)\\.cross_attn\\.q_linear\\.(.*)$': 'blocks.\\1.cross_attn.to_q.\\2', '^blocks\\.(\\d+)\\.cross_attn\\.kv_linear\\.(.*)$': 'blocks.\\1.cross_attn.kv_fused.\\2', '^blocks\\.(\\d+)\\.cross_attn\\.proj\\.(.*)$': 'blocks.\\1.cross_attn.to_out.\\2', '^blocks\\.(\\d+)\\.cross_attn\\.q_norm\\.(.*)$': 'blocks.\\1.cross_attn.q_norm.\\2', '^blocks\\.(\\d+)\\.cross_attn\\.k_norm\\.(.*)$': 'blocks.\\1.cross_attn.k_norm.\\2', '^blocks\\.(\\d+)\\.ffn\\.w1\\.(.*)$': 'blocks.\\1.ffn.w1.\\2', '^blocks\\.(\\d+)\\.ffn\\.w2\\.(.*)$': 'blocks.\\1.ffn.w2.\\2', '^blocks\\.(\\d+)\\.ffn\\.w3\\.(.*)$': 'blocks.\\1.ffn.w3.\\2', '^final_layer\\.adaLN_modulation\\.1\\.(.*)$': 'final_layer.adaln_linear.\\1', '^final_layer\\.norm_final\\.(.*)$': 'final_layer.norm.\\1', '^final_layer\\.linear\\.(.*)$': 'final_layer.proj.\\1'})(), reverse_param_names_mapping: dict = (lambda: {})(), lora_param_names_mapping: dict = (lambda: {})(), _supported_attention_backends: tuple = (lambda: (FLASH_ATTN, TORCH_SDPA))(), hidden_size: int = 4096, num_attention_heads: int = 32, num_channels_latents: int = 16, in_channels: int = 16, out_channels: int = 16, exclude_lora_layers: list[str] = (lambda: [])(), boundary_ratio: float | None = None, depth: int = 48, attention_head_dim: int = 128, patch_size: tuple[int, int, int] = (1, 2, 2), caption_channels: int = 4096, adaln_tembed_dim: int = 512, frequency_embedding_size: int = 256, mlp_ratio: int = 4, text_tokens_zero_pad: bool = True, enable_bsa: bool = False, bsa_params: dict | None = (lambda: {'sparsity': 0.9375, 'cdf_threshold': None, 'chunk_3d_shape_q': [4, 4, 4], 'chunk_3d_shape_k': [4, 4, 4]})())

Bases: DiTArchConfig

Architecture configuration for native LongCat Video DiT.

fastvideo.configs.models.dits.longcat.LongCatVideoConfig dataclass 
LongCatVideoConfig(arch_config: DiTArchConfig = LongCatVideoArchConfig(), prefix: str = 'longcat', quant_config: QuantizationConfig | None = None)

Bases: DiTConfig

Main configuration for LongCat Video DiT.

Functions
fastvideo.configs.models.dits.longcat.is_longcat_blocks 
is_longcat_blocks(n: str, m) -> bool

FSDP shard condition for LongCat transformer blocks.

Source code in fastvideo/configs/models/dits/longcat.py 
12
13
14
def is_longcat_blocks(n: str, m) -> bool:
    """FSDP shard condition for LongCat transformer blocks."""
    return "blocks" in n and str.isdigit(n.split(".")[-1])

fastvideo.configs.models.encoders

Classes

fastvideo.configs.models.encoders.Reason1ArchConfig dataclass 
Reason1ArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), architectures: list[str] = (lambda: ['Qwen2_5_VLForConditionalGeneration'])(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (FLASH_ATTN, TORCH_SDPA), output_hidden_states: bool = True, use_return_dict: bool = True, vocab_size: int = 152064, hidden_size: int = 3584, num_hidden_layers: int = 28, num_attention_heads: int = 28, pad_token_id: int = 151643, eos_token_id: int = 151645, text_len: int = 512, hidden_state_skip_layer: int = 0, decoder_start_token_id: int = 0, output_past: bool = True, scalable_attention: bool = True, tie_word_embeddings: bool = False, tokenizer_kwargs: dict[str, Any] = dict(), _fsdp_shard_conditions: list = (lambda: [])(), model_type: str = 'qwen2_5_vl', num_key_value_heads: int = 4, intermediate_size: int = 18944, bos_token_id: int = 151643, image_token_id: int = 151655, video_token_id: int = 151656, vision_token_id: int = 151654, vision_start_token_id: int = 151652, vision_end_token_id: int = 151653, vision_config: dict[str, Any] | None = None, rope_theta: float = 1000000.0, rope_scaling: dict[str, Any] | None = (lambda: {'type': 'mrope', 'mrope_section': [16, 24, 24]})(), max_position_embeddings: int = 128000, max_window_layers: int = 28, embedding_concat_strategy: str = 'mean_pooling', n_layers_per_group: int = 5, num_embedding_padding_tokens: int = 512, attention_dropout: float = 0.0, hidden_act: str = 'silu', initializer_range: float = 0.02, rms_norm_eps: float = 1e-06, use_sliding_window: bool = False, sliding_window: int = 32768, use_cache: bool = False, torch_dtype: str = 'bfloat16', _attn_implementation: str = 'flash_attention_2')

Bases: TextEncoderArchConfig

Architecture settings (defaults match Qwen2.5-VL-7B-Instruct).

fastvideo.configs.models.encoders.Reason1Config dataclass 
Reason1Config(arch_config: Reason1ArchConfig = Reason1ArchConfig(), prefix: str = '', quant_config: QuantizationConfig | None = None, lora_config: Any | None = None, is_chat_model: bool = False, tokenizer_type: str = 'Qwen/Qwen2.5-VL-7B-Instruct')

Bases: TextEncoderConfig

Reason1 text encoder config.

fastvideo.configs.models.encoders.T5LargeConfig dataclass 
T5LargeConfig(arch_config: TextEncoderArchConfig = T5LargeArchConfig(), prefix: str = 't5', quant_config: QuantizationConfig | None = None, lora_config: Any | None = None, is_chat_model: bool = False)

Bases: TextEncoderConfig

T5 Large configuration for your specific model.

Modules

fastvideo.configs.models.encoders.reason1

Config for Reason1 (Qwen2.5-VL) text encoder.

Classes
fastvideo.configs.models.encoders.reason1.Reason1ArchConfig dataclass 
Reason1ArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), architectures: list[str] = (lambda: ['Qwen2_5_VLForConditionalGeneration'])(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (FLASH_ATTN, TORCH_SDPA), output_hidden_states: bool = True, use_return_dict: bool = True, vocab_size: int = 152064, hidden_size: int = 3584, num_hidden_layers: int = 28, num_attention_heads: int = 28, pad_token_id: int = 151643, eos_token_id: int = 151645, text_len: int = 512, hidden_state_skip_layer: int = 0, decoder_start_token_id: int = 0, output_past: bool = True, scalable_attention: bool = True, tie_word_embeddings: bool = False, tokenizer_kwargs: dict[str, Any] = dict(), _fsdp_shard_conditions: list = (lambda: [])(), model_type: str = 'qwen2_5_vl', num_key_value_heads: int = 4, intermediate_size: int = 18944, bos_token_id: int = 151643, image_token_id: int = 151655, video_token_id: int = 151656, vision_token_id: int = 151654, vision_start_token_id: int = 151652, vision_end_token_id: int = 151653, vision_config: dict[str, Any] | None = None, rope_theta: float = 1000000.0, rope_scaling: dict[str, Any] | None = (lambda: {'type': 'mrope', 'mrope_section': [16, 24, 24]})(), max_position_embeddings: int = 128000, max_window_layers: int = 28, embedding_concat_strategy: str = 'mean_pooling', n_layers_per_group: int = 5, num_embedding_padding_tokens: int = 512, attention_dropout: float = 0.0, hidden_act: str = 'silu', initializer_range: float = 0.02, rms_norm_eps: float = 1e-06, use_sliding_window: bool = False, sliding_window: int = 32768, use_cache: bool = False, torch_dtype: str = 'bfloat16', _attn_implementation: str = 'flash_attention_2')

Bases: TextEncoderArchConfig

Architecture settings (defaults match Qwen2.5-VL-7B-Instruct).

fastvideo.configs.models.encoders.reason1.Reason1Config dataclass 
Reason1Config(arch_config: Reason1ArchConfig = Reason1ArchConfig(), prefix: str = '', quant_config: QuantizationConfig | None = None, lora_config: Any | None = None, is_chat_model: bool = False, tokenizer_type: str = 'Qwen/Qwen2.5-VL-7B-Instruct')

Bases: TextEncoderConfig

Reason1 text encoder config.

fastvideo.configs.models.encoders.t5
Classes
fastvideo.configs.models.encoders.t5.T5LargeArchConfig dataclass 
T5LargeArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = (lambda: [('.qkv_proj', '.q', 'q'), ('.qkv_proj', '.k', 'k'), ('.qkv_proj', '.v', 'v')])(), architectures: list[str] = (lambda: [])(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (FLASH_ATTN, TORCH_SDPA), output_hidden_states: bool = False, use_return_dict: bool = True, vocab_size: int = 32128, hidden_size: int = 0, num_hidden_layers: int = 0, num_attention_heads: int = 0, pad_token_id: int = 0, eos_token_id: int = 1, text_len: int = 512, hidden_state_skip_layer: int = 0, decoder_start_token_id: int = 0, output_past: bool = True, scalable_attention: bool = True, tie_word_embeddings: bool = False, tokenizer_kwargs: dict[str, Any] = dict(), _fsdp_shard_conditions: list = (lambda: [_is_transformer_layer, _is_embeddings, _is_final_layernorm])(), d_model: int = 1024, d_kv: int = 128, d_ff: int = 65536, num_layers: int = 24, num_decoder_layers: int | None = 24, num_heads: int = 128, relative_attention_num_buckets: int = 32, relative_attention_max_distance: int = 128, dropout_rate: float = 0.1, layer_norm_epsilon: float = 1e-06, initializer_factor: float = 1.0, feed_forward_proj: str = 'relu', dense_act_fn: str = '', is_gated_act: bool = False, is_encoder_decoder: bool = True, use_cache: bool = True, classifier_dropout: float = 0.0, dtype: str | None = None, gradient_checkpointing: bool = False, n_positions: int = 512, task_specific_params: dict | None = None)

Bases: T5ArchConfig

T5 Large architecture config with parameters for your specific model.

fastvideo.configs.models.encoders.t5.T5LargeConfig dataclass 
T5LargeConfig(arch_config: TextEncoderArchConfig = T5LargeArchConfig(), prefix: str = 't5', quant_config: QuantizationConfig | None = None, lora_config: Any | None = None, is_chat_model: bool = False)

Bases: TextEncoderConfig

T5 Large configuration for your specific model.

fastvideo.configs.models.vaes

Classes

fastvideo.configs.models.vaes.Cosmos25VAEConfig dataclass 
Cosmos25VAEConfig(arch_config: Cosmos25VAEArchConfig = Cosmos25VAEArchConfig(), load_encoder: bool = True, load_decoder: bool = True, tile_sample_min_height: int = 256, tile_sample_min_width: int = 256, tile_sample_min_num_frames: int = 16, tile_sample_stride_height: int = 192, tile_sample_stride_width: int = 192, tile_sample_stride_num_frames: int = 12, blend_num_frames: int = 0, use_tiling: bool = False, use_temporal_tiling: bool = False, use_parallel_tiling: bool = False, use_temporal_scaling_frames: bool = True, use_feature_cache: bool = True)

Bases: VAEConfig

Cosmos2.5 VAE config.

Modules

fastvideo.configs.models.vaes.base
Classes
fastvideo.configs.models.vaes.base.VAEConfig dataclass 
VAEConfig(arch_config: VAEArchConfig = VAEArchConfig(), load_encoder: bool = True, load_decoder: bool = True, tile_sample_min_height: int = 256, tile_sample_min_width: int = 256, tile_sample_min_num_frames: int = 16, tile_sample_stride_height: int = 192, tile_sample_stride_width: int = 192, tile_sample_stride_num_frames: int = 12, blend_num_frames: int = 0, use_tiling: bool = True, use_temporal_tiling: bool = True, use_parallel_tiling: bool = True, use_temporal_scaling_frames: bool = True)

Bases: ModelConfig

Functions
fastvideo.configs.models.vaes.base.VAEConfig.add_cli_args staticmethod 
add_cli_args(parser: Any, prefix: str = 'vae-config') -> Any

Add CLI arguments for VAEConfig fields

Source code in fastvideo/configs/models/vaes/base.py 
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
@staticmethod
def add_cli_args(parser: Any, prefix: str = "vae-config") -> Any:
    """Add CLI arguments for VAEConfig fields"""
    parser.add_argument(
        f"--{prefix}.load-encoder",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.load_encoder",
        default=VAEConfig.load_encoder,
        help="Whether to load the VAE encoder",
    )
    parser.add_argument(
        f"--{prefix}.load-decoder",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.load_decoder",
        default=VAEConfig.load_decoder,
        help="Whether to load the VAE decoder",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-min-height",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_min_height",
        default=VAEConfig.tile_sample_min_height,
        help="Minimum height for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-min-width",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_min_width",
        default=VAEConfig.tile_sample_min_width,
        help="Minimum width for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-min-num-frames",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_min_num_frames",
        default=VAEConfig.tile_sample_min_num_frames,
        help="Minimum number of frames for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-stride-height",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_stride_height",
        default=VAEConfig.tile_sample_stride_height,
        help="Stride height for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-stride-width",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_stride_width",
        default=VAEConfig.tile_sample_stride_width,
        help="Stride width for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.tile-sample-stride-num-frames",
        type=int,
        dest=f"{prefix.replace('-', '_')}.tile_sample_stride_num_frames",
        default=VAEConfig.tile_sample_stride_num_frames,
        help="Stride number of frames for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.blend-num-frames",
        type=int,
        dest=f"{prefix.replace('-', '_')}.blend_num_frames",
        default=VAEConfig.blend_num_frames,
        help="Number of frames to blend for VAE tile sampling",
    )
    parser.add_argument(
        f"--{prefix}.use-tiling",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.use_tiling",
        default=VAEConfig.use_tiling,
        help="Whether to use tiling for VAE",
    )
    parser.add_argument(
        f"--{prefix}.use-temporal-tiling",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.use_temporal_tiling",
        default=VAEConfig.use_temporal_tiling,
        help="Whether to use temporal tiling for VAE",
    )
    parser.add_argument(
        f"--{prefix}.use-parallel-tiling",
        action=StoreBoolean,
        dest=f"{prefix.replace('-', '_')}.use_parallel_tiling",
        default=VAEConfig.use_parallel_tiling,
        help="Whether to use parallel tiling for VAE",
    )

    return parser
fastvideo.configs.models.vaes.cosmos2_5vae

Cosmos 2.5 (Wan2.1-style) VAE config and checkpoint-key mapping.

Classes
fastvideo.configs.models.vaes.cosmos2_5vae.Cosmos25VAEArchConfig dataclass 
Cosmos25VAEArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), scaling_factor: float | Tensor = 0, temporal_compression_ratio: int = 4, spatial_compression_ratio: int = 8, _name_or_path: str = '', base_dim: int = 96, decoder_base_dim: int | None = None, z_dim: int = 16, dim_mult: tuple[int, ...] = (1, 2, 4, 4), num_res_blocks: int = 2, attn_scales: tuple[float, ...] = (), temperal_downsample: tuple[bool, ...] = (False, True, True), dropout: float = 0.0, is_residual: bool = False, in_channels: int = 3, out_channels: int = 3, patch_size: int | None = None, scale_factor_temporal: int = 4, scale_factor_spatial: int = 8, clip_output: bool = True, latents_mean: tuple[float, ...] = (-0.7571, -0.7089, -0.9113, 0.1075, -0.1745, 0.9653, -0.1517, 1.5508, 0.4134, -0.0715, 0.5517, -0.3632, -0.1922, -0.9497, 0.2503, -0.2921), latents_std: tuple[float, ...] = (2.8184, 1.4541, 2.3275, 2.6558, 1.2196, 1.7708, 2.6052, 2.0743, 3.2687, 2.1526, 2.8652, 1.5579, 1.6382, 1.1253, 2.8251, 1.916), param_names_mapping: dict[str, str] = (lambda: {'^conv1\\.(.*)$': 'quant_conv.\\1', '^conv2\\.(.*)$': 'post_quant_conv.\\1', '^encoder\\.conv1\\.(.*)$': 'encoder.conv_in.\\1', '^decoder\\.conv1\\.(.*)$': 'decoder.conv_in.\\1', '^encoder\\.head\\.0\\.gamma$': 'encoder.norm_out.gamma', '^encoder\\.head\\.2\\.(.*)$': 'encoder.conv_out.\\1', '^decoder\\.head\\.0\\.gamma$': 'decoder.norm_out.gamma', '^decoder\\.head\\.2\\.(.*)$': 'decoder.conv_out.\\1'})())

Bases: VAEArchConfig

Functions
fastvideo.configs.models.vaes.cosmos2_5vae.Cosmos25VAEArchConfig.map_official_key staticmethod 
map_official_key(key: str) -> str | None

Map a single official checkpoint key into FastVideo key space.

Source code in fastvideo/configs/models/vaes/cosmos2_5vae.py 
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
@staticmethod
def map_official_key(key: str) -> str | None:
    """Map a single official checkpoint key into FastVideo key space."""

    def map_residual_subkey(prefix: str, sub: str) -> str | None:
        if re.match(r"^residual\.0\.gamma$", sub):
            return f"{prefix}.norm1.gamma"
        m = re.match(r"^residual\.2\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.conv1.{m.group(1)}"
        if re.match(r"^residual\.3\.gamma$", sub):
            return f"{prefix}.norm2.gamma"
        m = re.match(r"^residual\.6\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.conv2.{m.group(1)}"
        m = re.match(r"^shortcut\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.conv_shortcut.{m.group(1)}"
        return None

    def map_attn_subkey(prefix: str, sub: str) -> str | None:
        if re.match(r"^norm\.gamma$", sub):
            return f"{prefix}.norm.gamma"
        m = re.match(r"^to_qkv\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.to_qkv.{m.group(1)}"
        m = re.match(r"^proj\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.proj.{m.group(1)}"
        return None

    def map_resample_subkey(prefix: str, sub: str) -> str | None:
        m = re.match(r"^resample\.1\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.resample.1.{m.group(1)}"
        m = re.match(r"^time_conv\.(weight|bias)$", sub)
        if m:
            return f"{prefix}.time_conv.{m.group(1)}"
        return None

    m = re.match(r"^conv1\.(weight|bias)$", key)
    if m:
        return f"quant_conv.{m.group(1)}"
    m = re.match(r"^conv2\.(weight|bias)$", key)
    if m:
        return f"post_quant_conv.{m.group(1)}"
    m = re.match(r"^(encoder|decoder)\.conv1\.(weight|bias)$", key)
    if m:
        return f"{m.group(1)}.conv_in.{m.group(2)}"
    m = re.match(r"^(encoder|decoder)\.head\.0\.gamma$", key)
    if m:
        return f"{m.group(1)}.norm_out.gamma"
    m = re.match(r"^(encoder|decoder)\.head\.2\.(weight|bias)$", key)
    if m:
        return f"{m.group(1)}.conv_out.{m.group(2)}"

    m = re.match(r"^(encoder|decoder)\.middle\.0\.(.*)$", key)
    if m:
        return map_residual_subkey(f"{m.group(1)}.mid_block.resnets.0",
                                   m.group(2))
    m = re.match(r"^(encoder|decoder)\.middle\.1\.(.*)$", key)
    if m:
        return map_attn_subkey(f"{m.group(1)}.mid_block.attentions.0",
                               m.group(2))
    m = re.match(r"^(encoder|decoder)\.middle\.2\.(.*)$", key)
    if m:
        return map_residual_subkey(f"{m.group(1)}.mid_block.resnets.1",
                                   m.group(2))

    m = re.match(r"^encoder\.downsamples\.(\d+)\.(.*)$", key)
    if m:
        idx = int(m.group(1))
        sub = m.group(2)
        if sub.startswith("residual.") or sub.startswith("shortcut."):
            return map_residual_subkey(f"encoder.down_blocks.{idx}", sub)
        if sub.startswith("resample.") or sub.startswith("time_conv."):
            return map_resample_subkey(f"encoder.down_blocks.{idx}", sub)
        return None

    m = re.match(r"^decoder\.upsamples\.(\d+)\.(.*)$", key)
    if m:
        uidx = int(m.group(1))
        sub = m.group(2)

        if uidx in (0, 1, 2):
            block_i, res_i = 0, uidx
        elif uidx == 3:
            block_i, res_i = 0, None
        elif uidx in (4, 5, 6):
            block_i, res_i = 1, uidx - 4
        elif uidx == 7:
            block_i, res_i = 1, None
        elif uidx in (8, 9, 10):
            block_i, res_i = 2, uidx - 8
        elif uidx == 11:
            block_i, res_i = 2, None
        elif uidx in (12, 13, 14):
            block_i, res_i = 3, uidx - 12
        else:
            return None

        if res_i is None:
            return map_resample_subkey(
                f"decoder.up_blocks.{block_i}.upsamplers.0",
                sub,
            )

        return map_residual_subkey(
            f"decoder.up_blocks.{block_i}.resnets.{res_i}",
            sub,
        )

    return None
fastvideo.configs.models.vaes.cosmos2_5vae.Cosmos25VAEConfig dataclass 
Cosmos25VAEConfig(arch_config: Cosmos25VAEArchConfig = Cosmos25VAEArchConfig(), load_encoder: bool = True, load_decoder: bool = True, tile_sample_min_height: int = 256, tile_sample_min_width: int = 256, tile_sample_min_num_frames: int = 16, tile_sample_stride_height: int = 192, tile_sample_stride_width: int = 192, tile_sample_stride_num_frames: int = 12, blend_num_frames: int = 0, use_tiling: bool = False, use_temporal_tiling: bool = False, use_parallel_tiling: bool = False, use_temporal_scaling_frames: bool = True, use_feature_cache: bool = True)

Bases: VAEConfig

Cosmos2.5 VAE config.